Environmental Science Processes & Impacts
View Article Online PAPER View Journal | View Issue
Response of discharge, TSS, and E. coli to rainfall events in urban, suburban, and rural watersheds Cite this: Environ. Sci.: Processes Impacts ,2014,16,2313 H. J. Chena and H. Chang*b
Understanding dominant processes influencing microorganism responses to storm events aids in the development of effective management controls on pathogen contamination in surface water so that they are suitable for water supply, recreation, and aquatic habitat. Despite the urgent needs at present, numerous facets of microbial transport and fate are still poorly understood. Using correlation and multiple regression combined with spatial analyses, this paper evaluates the relationship between antecedent precipitation and discharge, TSS, and E. coli concentrations, and examines correlations between E. coli and TSS, as well as whether and how those relationships change along an urban and rural gradient. The urban watershed exhibited a faster and stronger response of streamflow, TSS, and E. coli to precipitation mainly due to its higher degree of imperviousness. In general, TSS was significantly correlated with E. coli concentrations, which linearly decreased as % developed area increased, with large variation in regions with a high percentage of development, implying the more complex stormwater infrastructure and more variable pollutant sources of E. coli in the urban watershed. Seasonal differences for E. coli were noted. Specifically, summer showed a higher level of E. coli, which might be attributed to the higher temperature since E. coli is more likely to persist and grow in a warmer environment. Further multiple linear regression analyses showed the best E. coli prediction result for the largest, suburban watershed, using antecedent precipitation, TSS, and temperature as independent variables. The models are capable of explaining 60% and 50% of the variability in the E. coli Received 12th June 2014 concentration for the dry and wet season, respectively. The study not only provides more detailed and Accepted 14th July 2014 accurate characterization of the storm-period response of E. coli across an urban and rural gradient, but DOI: 10.1039/c4em00327f also lays a foundation for predicting the concentration of E. coli in practice, potentially suggesting rsc.li/process-impacts effective watershed management decisions. Published on 14 July 2014. Downloaded by University of California - Davis 21/01/2016 01:40:41. Environmental impact E. Coli is an increasing concern in many watersheds. Since elevated levels of E. coli can negatively affect water supply, recreation, and aquatic habitat, identifying possible sources and transport of E. coli has been a primary concern in environmental sciences and management. Here we investigated how the sources and transport mechanisms might differ across different levels of urban development during storm events. Identifying potential factors that affect E. coli concen- trations can not only help control sources, but also provide information to better predict changing levels of E. coli as they relate to other pollutants or mete- orological factors. We attempt to unravel the dynamics of E. coli concentration using a combination of meteorological and landscape factors during storm events.
Introduction possibility that the water has been polluted by feces associated with pathogens. Simulation models can play an important role Pathogens are the number one cause of impairment for Clean in the assessment and management of microbial contamina- Water Act (CWA) Section 303(d) listed waters in the USA and tion. However, the development of such a model requires an pose a signicant threat on human health and quality of life.1 accurate understanding of the transport, build-up and persis- Fecal coliform bacteria, particularly E. coli, are oen selected as tence of microorganisms in the catchment system.2 critical biological indicators of the presence of pathogens. The Recent studies have shown that levels of E. coli dramatically – higher the level of E. coli density in water bodies, the greater increased in response to storm events,2 4 indicating that wash- off models, in which stormwater runoff serves as a contributor to pollutants in surface waters, will partially explain the aDepartment of Land, Air and Water Resources, University of California, Davis. One considerable inter-event variability in E. coli concentrations. Shields Avenue, Davis, CA 95616-5270, USA Moreover, a signicant correlation was observed between E. coli bDepartment of Geography, Portland State University, PO Box 751 – Geog, Portland, 5–7 Oregon 97207-0751, USA. E-mail: [email protected] and preceding rainfall events. These ndings suggest that the
This journal is © The Royal Society of Chemistry 2014 Environ. Sci.: Processes Impacts,2014,16,2313–2324 | 2313 View Article Online Environmental Science: Processes & Impacts Paper
antecedent rainfall conditions are likely to impact both the variable. The R2 values of their models ranged from 0.67 to amount of water and energy available for E. coli transport and 0.84.22 Hamilton and Luffman (2009)7 have achieved relative the amount of moisture present in a watershed that is critical success in using multiple linear regression analysis (R2 ¼ 0.565) for E. coli survival.6 In a study of Stock Creek, Tennessee, to predict the concentration of E. coli using precipitation, however, there was no statistically signicant correlation discharge, and turbidity as predictors. In a study conducted between E. coli and precipitation.8 Therefore, a more compre- along tributaries of Tualatin River, Oregon, discharge and hensive understanding of the relationship between rainfall and turbidity were selected as predictors in the regression equa- the presence of E. coli across different types of watersheds is tion.18 The study showed moderately successful prediction for E. needed, which is critical for managing water systems so that coli bacteria with reasonably high R2 values (0.586–0.713). water managers are able to provide potable water and water Although several studies have reported that antecedent precip- suitable for recreation and aquatic habitat. itation, stream discharge, sediment density, and water Apart from inter-event variation, E. coli concentrations also temperature could serve as potential predictors for E. coli signicantly vary by season.6,9 Temperature, which is known to concentrations, it is still unclear how that pattern would change inuence the survival of E. coli, has been used to help explain across an urban and rural gradient. the variation of E. coli levels between dry and wet periods. Many The objective of this study was to develop more detailed and studies have shown that die off rates for E. coli are higher as accurate characterization of the storm-period response of E. coli temperature increases, which could be attributed to stronger in urban, suburban, and rural watersheds, potentially allowing sunlight radiation in warmer seasons.10–12 However, reverse effective management controls on pathogen contamination in trends were observed in several studies where E. coli levels were these watersheds. Specically, the goals are to: higher during summer seasons. These contradictory results are (1) Evaluate the relationship between antecedent precipita- likely consequences of the summer seasons having the warmest tion and the three other parameters: discharge, TSS, and E. coli temperatures and less streamow, which corresponds to higher concentrations. growth and survival rates of E. coli bacteria9,13 and less dilution (2) Determine the correlations between E. coli and TSS, as effect.5 In addition, higher E. coli concentrations in urban areas well as whether and how those relationships change along an may also be attributed to the increase in animal and human urban and rural gradient. activities during the warm season.14 (3) Investigate the variation of discharge, TSS, temperature The majority of E. coli in aquatic systems is also associated and E. coli between dry and wet seasons. with sediments, and these associations increase the survival (4) Determine the correlation between discharge, TSS, rate of fecal bacteria relative to those in the water column.15,16 temperature, E. coli and land use type. Positive correlations between total suspended solids (TSS) and (5) Construct regression models to predict the concentra- E. coli concentrations have been observed by Anderson and tions of E. coli using antecedent precipitation, stream Rounds (2003)13 and Hamilton & Luffman (2009)7 for highly discharge, TSS and temperature as easy-to-measure proxies, and urbanized watersheds (Fanno Creek watershed, Oregon; Little compare model signicance and parameters among the three River watershed, Tennessee), and Muirhead et al. (2004)17 for watersheds. articial ood events in pasture land. These ndings indicate Published on 14 July 2014. Downloaded by University of California - Davis 21/01/2016 01:40:41. that E. coli were either transported to stream bound to partic- Study area ulate matter, adsorbed onto resuspended stream bed particles, or they had an affinity for sediments in water.18 Therefore, The sample sites were located in the Portland Metropolitan area sediments have the potential to serve as a surrogate for E. coli (Fig. 1), which comprises Clackamas, Columbia, Multnomah, concentrations. In two other studies, however, only a weak Washington, and Yamhill Counties. Climates of these counties relationship between E. coli and TSS was observed in articial are heavily inuenced by the Pacic Ocean, and are character- ood events in the northern England19 and on marsh lands at ized by mild, humid winter and hot, dry summer. Future Texas coast.20 Given the large spatial variation of the correlation climate is projected to be drier and hotter in summer and wetter between E. coli and TSS, site-specic landscape patterns within in winter,23 increasing the probability of droughts24 and a given watershed may have important impacts on the E. coli- oods.25 Being one of the fastest growing metropolitan areas of sediment relationship, which has not been fully characterized. the USA,26 ongoing urban development is posing water quality Regression analysis has been applied to predict E. coli concerns, particularly in the urban–rural fringe area.27 The land concentrations and determine the nature and causes of its use varies from heavily developed urban areas in the middle variability. Several studies have reported that E. coli concen- part to rural and agricultural in the eastern and far western trations are strongly related to antecedent precipitation, sedi- extents. We selected three watersheds that represent a gradient ments, streamow characteristics, temperature, and of urban development. They are Fanno Creek, Johnson Creek, season.7,18,21 Linear regression is one of the most commonly and Balch Creek. We refer to Fanno Creek as urban, Johnson used statistical methods in water quality research. Rasmussen Creek as mixed, and Balch Creek as forested, according to their and others (2008)22 have developed simple linear regression land use patterns. models to perform real time prediction for 19 constituents, Fanno Creek is a 24 km tributary of the Tualatin River and including fecal coliforms in streams of Johnson County, ows west from its headwaters in Hillsdale to its conuence northeast Kansas, using turbidity as the only explanatory with the Tualatin River near Durham, draining an area of
2314 | Environ. Sci.: Processes Impacts,2014,16,2313–2324 This journal is © The Royal Society of Chemistry 2014 View Article Online Paper Environmental Science: Processes & Impacts
Fig. 1 Study area and location of sampling sites.
83 km2. Fanno Creek was chosen as the study site because it ows west and drains an area of approximately 137.6 km2 with serves as a representative for highly urbanized watersheds. 180 000 residents. The Johnson Creek watershed is moderately Covering several major cities in Washington County including developed with approximately 50% of the watershed urbanized, Portland, Beaverton, Tigard and Durham, the Fanno Creek mostly within the urban growth boundary in the lower and watershed is highly developed with 84% urban land use.28 middle reaches of the catchment. There are quite a few creek- Fanno Creek is also a typical stream system that is impaired by side parks along the Johnson Creek, including sports elds, stormwater runoff from existing point sources and develop- picnic areas, and trails. For example, the 20 mile spring water ment29 and is listed on the Oregon's 303(d) list (Table 1) due to corridor, which runs in parallel with the Johnson Creek between high levels of ammonia, nutrients, total solids, and E. coli which the mouth of the Johnson Creek and the mid-point of the exceeded the water quality standard in 50% of samples in mainstem Johnson Creek, has been a popular place for running, 29 Published on 14 July 2014. Downloaded by University of California - Davis 21/01/2016 01:40:41. summer and 25% during winter. Although heavily polluted, cycling, walking, and other human activities. Kids and pets play the creek supports aquatic life in upper reaches and passes with the creek water in these recreational areas adjacent to the through or close to 14 parks in several jurisdictions, serving a creek. The upper part is primarily used for rural and agricultural recreational function for nearby residents. Despite tremendous lands. The Johnson Creek fails to meet the state health stan- efforts in stormwater management and increase in public dards for contact recreation because of high levels of E. coli awareness, no remarkable improvement in water quality in bacteria.31 E. coli contamination is a major concern for both Fanno Creek was observed,30 primarily owing to the high stream health and property sales price. A recent study shows proportion of impervious surfaces as well as the loss of riparian that higher E. coli concentrations have negative inuence on areas in the watershed. The locations and land cover of the home sales price.32 sample sites are shown in Fig. 1 and summarized in Tables 2 The Balch Creek is a 5.6 km tributary of the Willamette River. and 3. It drains a small basin of 9.1 km2 at the central part of the Located on the eastern side of the Portland Metropolitan Portland Metropolitan area. Originating from the crest of the region, the Johnson Creek is a 40.2 km tributary of the Will- Tualatin Mountains, the stream ows east through the Macleay amette River. Beginning at Clackamas County, east of Boring, it Park section of Forest Park, a large municipal park in Portland.
Table 1 Oregon's 303(d) list – Fanno Creek
Parameter related Season Benecial uses Status
E. coli Fall winter spring Water contact recreation Cat 4A: Water quality limited, TMDL approved E. coli Summer Water contact recreation Cat 4A: Water quality limited, TMDL approved Temperature Summer Salmonid sh rearing; anadromous sh passage
This journal is © The Royal Society of Chemistry 2014 Environ. Sci.: Processes Impacts,2014,16,2313–2324 | 2315 View Article Online Environmental Science: Processes & Impacts Paper
Table 2 Summary of rainfall and discharge gages
Rainfall Discharge
Watershed Station name Long. Lat. USGS site number Long. Lat.
Fanno Sylvania PCC 122.73 45.44 14206900 122.73 45.49 14206950 122.75 45.40 Johnson Kelly School 122.57 45.47 14211550 122.64 45.45 Hayney 122.64 45.46 14211500 122.51 45.48 Pleasant Valley School 122.48 45.46 14211400 122.42 45.49 14211499 122.5 45.48 Balch Yeon 122.71 45.55 NA
Aer entering a pipe at the lower end of the park, the creek (Hydrological Data Retrieval and Alarm) Rainfall Network remains underground until reaching the river. Most parts of the operated and maintained by the City of Portland's Bureau of Balch Creek watershed remain as a rural and open area as well Environmental Services (BES). Daily stream discharge data were as forestry (Fig. 1). Only a tiny part of the watershed is used for obtained from USGS (the United States Geological Survey). residential and commercial land uses. Water quality data such as TSS, stream temperature, and E. coli For the purposes of this analysis, the Fanno Creek is the concentrations were obtained from the datasets maintained by most developed watershed in the Portland Metropolitan area, BES and Clean Water Services, the managing agency respon- which has more than 84% of urban land use and little agri- sible for water and sewer management in Washington County, cultural or forestry usage (Table 3). The Johnson Creek has Oregon. The summaries of monitoring stations are listed in approximately 50% of urban area, and yet retains a certain Tables 2 and 3. amount of forested and agricultural land use (Table 3). The Data were extracted from July 2002 to June 2010. Precipita- Balch Creek is the only watershed of the three with signicant tion data were recorded in millimeter (mm) as a 24 hour total. rural and forest coverage, and very little residential and busi- The three-day, ve-day, and seven-day antecedent precipitation ness usage (Table 3). data were calculated by adding up the prior precipitation correspondingly. Data for streamow were recorded through an automatic system at one-minute intervals, transmitted from Method each station at intervals of 3–6 hours, and then loaded onto the Data sources USGS computer system. Daily discharge measurements were We obtained data from three primary sources (Table 4). Daily then calculated as the average of discharge measurements over 33 precipitation data came from the City of Portland HYDRA a 24 hour period and are reported in cubic meter per second.
Published on 14 July 2014. Downloaded by University of California - Davis 21/01/2016 01:40:41. Table 3 Summary of water quality monitor stations
Watershed Station name Long. Lat. Slope (degree) %IMP %Dev %Forest
Fanno N Ash 122.74 45.46 16.7 39.4 100.0 0.0 S Ash 122.74 45.45 13.5 39.0 100.0 0.0 Main 3975 122.72 45.49 14.6 55.0 100.0 0.0 Main 4916 122.73 45.49 5.2 56.1 100.0 0.0 Main 6900 122.75 45.49 5.8 91.0 100.0 0.0 Pendleton 122.74 45.49 7.2 41.5 100.0 0.0 Vermont 122.75 45.48 20.0 28.7 100.0 0.0 Woods 122.75 45.47 18.2 14.0 100.0 0.0 Durham 122.75 45.40 11.2 21.8 92.2 0.0 Johnson Crystal 122.64 45.47 1.0 61.0 100.0 0.0 SE Regner 122.42 45.49 11.2 9.4 33.1 20.4 SE Umatilla 122.64 45.46 11.6 51.8 97.2 1.8 East of Johnson 122.60 45.46 12.4 44.4 88.1 9.5 SE 92 122.57 45.47 17.0 86.0 100.0 0.0 SE 158 122.50 45.48 0.3 30.7 69.1 24.3 SW Pleasant 122.48 45.49 24.7 43.0 66.1 33.9 SE Hogan 122.41 45.48 13.2 14.1 100.0 0.0 SE 159 122.50 45.48 10.5 10.7 37.3 28.2 Balch Thompson 122.74 45.53 16.9 4.0 24.1 75.1 East of Bones 122.73 45.53 21.8 44.4 88.1 9.5 Cornell 122.73 45.53 12.4 14.0 63.6 36.4 L Macleay 122.71 45.54 9.6 36.7 100.0 0.0
2316 | Environ. Sci.: Processes Impacts,2014,16, 2313–2324 This journal is © The Royal Society of Chemistry 2014 View Article Online Paper Environmental Science: Processes & Impacts
Table 4 Summary of data sources
Data Source
Precipitation City of Portland HYDRA (Hydrological Data Retrieval and Alarm) Rainfall Network Stream discharge USGS (the United States Geological Survey) Water quality City of Portland's Bureau of Environmental Services (BES) and Clean Water Services, Oregon Land cover, imperviousness estimate, and digital elevation USGS (the United States Geological Survey) model Stream network Metro's RLIS (Regional Land Information System)
All discharge data used in the analyses have been quality- precipitation and the three other parameters respectively: assured and approved for publication by USGS. discharge, TSS, and E. coli. The one-day, three-day, ve-day, and Water quality data were collected monthly, on site for water seven-day antecedent precipitation were all included in order to temperature, total suspended solids (TSS), and E. coli concen- fully consider the rainfall events that last longer than one day.7 trations. Laboratory analyses were conducted by the Clean Second, the cross-correlations between E. coli and discharge, Water Services water-quality laboratory in Hillsboro, Oregon, TSS, and water temperature were determined to further char- using protocols described in Standard Methods for the Exami- acterize the possible association between E. coli and other nation of Water and Wastewater, 2005.34 TSS was measured environmental variables. according to SM 2540 D. Well-mixed samples were ltered The dry–wet seasonal differences in water quality were then through a weighed standard glass-ber lter. The retained investigated. Again, our non-normally distributed data only residue was dried to a constant weight at 103 to 105 C, and TSS allow us to perform the Mann–Whitney U test, which is used to was calculated as the increase in weight of the lter. Tempera- compare differences between two independent groups without ture measurements were made with a mercury-lled Celsius assumption about normality. The dry and wet seasons were thermometer (SM 2550 B). E. coli concentrations were measured assigned as follows: dry season ¼ May–October; wet season ¼ using a 9223B Enzyme Substrate Test. Samples were mixed with November (of the preceding year)–April, following a previous enzyme substrates and incubated at 35 0.5 C. Beta-glucu- study in the study area.36,37 ronidase, an enzyme produced by E. coli, was detected by The Spearman ranking correlation between water quality hydrolysis of the uorescent substrate MUG (4-methyl- and landscape variables was performed in both dry and wet umbelliferyl-beta-D-glucuronide). Hydrolyzed MUG was seen as seasons, across the Fanno, Johnson, and Balch Creek water- blue uorescence when viewed under long-wavelength (366 nm) sheds. Specically, we examined the relationship between E. coli ultraviolet light, indicating a positive test for E. coli. At least 10% and TSS, with %DEV, %IMP and %Forest across all delineated of all samples were analyzed independently in duplicate, which subwatersheds. Correlation results are considered signicant at agreed within 5% of their average values.34 the 0.05 level. Land cover and imperviousness estimate layers were Published on 14 July 2014. Downloaded by University of California - Davis 21/01/2016 01:40:41. provided by the US Geological Survey (http://www.mrlc.gov/ nlcd06_data.php). A thirty meter digital elevation model was also obtained from USGS. A stream network layer was obtained Regression analysis from Metro's RLIS (Regional Land Information System). Two types of multiple regression analysis were used. First, multiple linear regression was applied to determine the Spatial analysis response of a correlation between TSS and E. coli concentrations Each study catchment area was derived using ArcMap 10.1 to land cover types. The percentage of imperviousness (%IMP), geographic information system soware. The subwatershed percentage of Development (%DEV), and %Forest were selected drained by each location was determined using the ArcHydro as the independent variables, and seasons (dry and wet) as well extension and digital elevation models produced by the US as watersheds (Fanno and Johnson) were used as grouping Geologic Survey at 30 m resolution (http://ned.usgs.gov/). Land variables. The Balch Creek watershed was excluded from the parcel polygons within each delineated drainage basin were analysis due to the lack of signi cance data points. The full selected and summarized by the land use type. Those types model was constructed as: utilized in this analysis are: developed land (DEV), forest, and Cor ¼ season + watershed + X + season X + watershed X impervious surfaces (IMP). + season watershed,
Statistical analysis where Cor stands for the correlation between TSS and E. coli, According to the Shapiro–Wilk test, our dataset is not normally and X stands for %IMP or %DEV or %Forest. Variables selection distributed. Hence, the Spearman's rank correlation method— was performed using stepwise AIC in soware R. We would which makes no assumptions about the distribution of the expect decrease in correlation strength as %IMP and %DEV go data35—was applied to examine the correlations between up, and %Forest goes down. The R2 value was reported that
This journal is © The Royal Society of Chemistry 2014 Environ. Sci.: Processes Impacts,2014,16, 2313–2324 | 2317 View Article Online Environmental Science: Processes & Impacts Paper
represents the proportion of variation in the response variable where more time is needed for soil to be saturated, and then explained by the tted regression line.38 runoff can start. A similar result was found in small Pennsyl- Second, we employed multiple linear regression to model the vania watersheds.41 This may also suggests a signicant base determinants of the concentrations of E. coli based on previous ow component to Johnson Creek discharge from groundwater correlation analyses. Precipitation, discharge, TSS, and or septic systems in rural areas or old residential areas. temperature are the likely candidates for explaining the concentration of E. coli as these variables could strongly inu- ence the transport, build-up, and survival of E. coli. Log trans- Correlation between TSS and antecedent precipitation formation, which can provide better homoscedasticity and result in a more symmetric dataset with normal residuals,39 was TSS was more strongly correlated with the one-day and three- performed on E. coli concentrations. This approach has been day than the seven-day antecedent precipitation for dry and wet used successfully for E. coli concentrations and other selected seasons in the Fanno Creek (Table 5). This fast response relative variables in streams in Oregon.13 Aer natural log trans- to discharge may indicate the existence of a rst ush effect, formation, our E. coli data are able to t into a normal distri- which was also observed in urban stormwater in Raleigh, North bution, proving the applicability of regression analysis. The Carolina, U.S.3 In other words, most sediments were delivered goodness-of-t of predictive models was assessed with diag- to the stream at the initial stage (in terms of runoff volume) of nostics statistics, including residual plots and the R2. ANOVA the rainfall events, meaning that the sources were either near tables were also employed to identify the signicance of our streams or from top soils that had been accumulated between models at the 95 percent condence interval (p ¼ 0.05). rainfall events. The assumption seems to be reasonable since all monitoring stations are located in either high- or medium- Results and discussion developed areas in the Fanno Creek watershed, and most are close to parks and other green spaces where people typically Correlation between discharge and antecedent precipitation accompany dogs or other pets. amount The Johnson Creek showed a slow response in TSS with the A strongest signicant positive correlation was observed strongest correlation with the ve-day antecedent precipitation between stream discharge and the three-day antecedent in the wet season, and no signicant correlation in the dry precipitation in the Fanno Creek watershed for both dry and wet season (Table 5). The results would accord with the previous seasons (r ¼ 0.547, 0.759 respectively, Table 5). This relatively discussion regarding streamow; since most precipitation made fast response of streamow to precipitation is not surprising in its way down through the soil into groundwater, and the rela- highly urbanized watersheds with a steep slope like the Fanno tively small amount of surface runoff was not be able to carry Creek watershed (Table 3), where higher impervious surfaces sediments into the stream systems, resulting in little TSS associated with high-density development are likely to promote response to rainfall events. This could also be related to the hydraulic efficiency, resulting in a faster response of streamow dominant contribution of baseow during the dry period, which to precipitation events.40 might attenuate or overshadow the pollutant trends in storm- For discharge–precipitation correlation, the Johnson Creek water. Moreover, some sections of the Johnson Creek are
Published on 14 July 2014. Downloaded by University of California - Davis 21/01/2016 01:40:41. was more highly correlated with the seven-day antecedent armored from the WPA (Works Progress Administration) era. precipitation than with the 3 day or 5 day antecedent precipi- The consolidated and armored river bank tends to have less bank tation for both seasons (Table 5). This slow response could be erosion and consequently fewer storm-associated sediments. attributed to the large size, at topography, and elongated No signicant rainfall correlation with TSS was observed in shape of the Johnson Creek watershed. Moreover, unlike the the Balch Creek during both seasons. Since the Balch Creek is Fanno Creek that is mostly covered by urban land use (Table 3), the least urbanized stream system, possible explanations could land cover upstream of monitoring stations in the Johnson be that sediment sources are scarce or soils can effectively Creek watershed is dominated by agricultural and rural lands absorb sediments. Also, similar to the Johnson Creek, the
Table 5 Spearman's correlation coefficient in the Fanno Creek, Johnson Creek and Balch Creek watersheds (nFanno,dry ¼ 66, nFanno,wet ¼ 68; a nJohnson,dry ¼ 52, nJohnson,wet ¼ 46; nBalch,dry ¼ 48, nBalch,wet ¼ 35)
Discharge TSS E. coli
Fanno Johnson Fanno Johnson Balch Fanno Johnson Balch
Dry Wet Dry Wet Dry Wet Dry Wet Dry Wet Dry Wet Dry Wet Dry Wet
Precip-1 0.525* 0.720* 0.480* ns ns 0.490* ns ns ns ns 0.498* 0.539* ns 0.658* ns ns Precip-3 0.547* 0.759* 0.585* 0.585* ns 0.497* ns 0.753* ns ns 0.393* 0.426* ns 0.615* ns 0.574* Precip-5 0.527* 0.709* 0.560* 0.700* ns 0.419* ns 0.796* ns ns 0.391* 0.350* ns 0.399* ns 0.658* Precip-7 0.547* 0.681* 0.596* 0.723* ns 0.416* ns 0.692* ns ns 0.322* ns ns ns ns 0.551* a ns: non-signicant; * signicant at the 0.05 level; the highest coefficient values are in bold type.
2318 | Environ. Sci.: Processes Impacts,2014,16,2313–2324 This journal is © The Royal Society of Chemistry 2014 View Article Online Paper Environmental Science: Processes & Impacts
higher surface permeability might generate too scarce runoff to Our results are largely consistent with previous ndings wash off pollutants into streams. from Anderson and Rounds (2003)13 and Hamilton & Luffman (2009)7 for highly urbanized watersheds in Oregon and eastern Correlation between E. coli and antecedent precipitation Tennessee, U.S., respectively; both determined that E. coli bacteria was positively correlated with turbidity, indicating that In all three watersheds, signicant correlation was observed E. coli were either transported to stream bound to particulate (Table 5) between E. coli and preceding rainfall events, at least matter, adsorbed onto resuspended stream bed particles, or during the wet season, which accords with previous studies.5–7 they had an affinity for sediments in water.18 It is reasonable to This is probably because the antecedent rainfall conditions suggest that the majority of E. coli sources lie close to the affect both the amount of water and energy available for E. coli Johnson Creek, especially considering spring corridor trails transport and the amount of moisture present in a watershed parallel to the creek in the middle and lower section of the creek that is critical for E. coli survival.6 Similar to TSS, E. coli in the and the agricultural activities in riparian areas of the upper Fanno Creek was most strongly correlated with the one-day section of the creek that apply a large amount of manures. It antecedent precipitation, which might also indicate a rst ush should also be noticed that E. coli is an indicator of recent effect on E. coli. In the Johnson Creek, however, E. coli showed a pollution which could attenuate the correlations between E. coli different response, compared to TSS response to precipitation, concentrations and the seven-day antecedent precipitation. with the strongest correlation with the one-day antecedent Samples and analyses for indicators of old pollution (e.g., precipitation during the wet season (r ¼ 0.658). Enterococcus faecalis) are needed for further assessment. However, one cannot simply infer that E. coli is transported E. coli Correlation between and TSS with suspended sediments, especially given the distinct As shown in Fig. 2, E. coli is generally positively associated with responses of TSS and E. coli to storm events in the Johnson TSS, with higher E. coli and lower TSS concentrations in the dry Creek. The bifurcation in the E. coli versus TSS plot of the Balch season than that in the wet season. The correlations are all Creek watershed in the wet season (Fig. 2) could suggest two statistically signicant in three watersheds in the dry season, different TSS sources with different amounts of E. coli associ- while the correlation for the Balch Creek is not statistically ated with them. One possible source of sediment is from near signicant in the wet season (Table 6). While the Balch Creek streams such as the stream bank or stream bed, and the other has the highest correlation in the dry season, the Johnson Creek potential source is from distant areas such as upstream areas or has the highest correlation in the wet season. water delivered from storm pipes.
Seasonal differences in water quality According to the Mann–Whitney U test, discharge, temperature, and E. coli showed signicant dry–wet seasonal differences across Fanno, Johnson, and Balch Creek watersheds (p ¼ 0.01, 2-tailed). Not surprisingly, higher discharge and lower temper- Published on 14 July 2014. Downloaded by University of California - Davis 21/01/2016 01:40:41. ature were observed during the wet season. E. coli concentra- tions are signicantly higher in the dry season than in the wet season (Fig. 3). This is a likely consequence of higher ows and more frequent washout of stored bacteria in the wet season.42 The increase in animal and human activities during the warm season could also lead to higher E. coli concentrations in streams. Most occurrences of elevated E. coli levels in urban watersheds originate from sources such as domestic pet waste.14 People typically walk their pets more oen during the warmer season, and therefore increasing the probability of feces contamination.
Table 6 Spearman's rank correlation coefficient between E. coli and TSSa
Dry Wet All season
Fanno 0.246* (n ¼ 66) 0.321* (n ¼ 68) 0.323* (n ¼ 134) Johnson 0.359* (n ¼ 52) 0.423* (n ¼ 46) 0.299* (n ¼ 98) Balch 0.380* (n ¼ 48) 0.222 (n ¼ 35) 0.223* (n ¼ 83) Fig. 2 Relationship between E. coli and TSS in three watersheds for (a) the dry season and (b) the wet season; all correlation results are a *: signicant at the 0.05 level; the highest coefficient values are in bold significant at the 0.05 level. type.
This journal is © The Royal Society of Chemistry 2014 Environ. Sci.: Processes Impacts,2014,16, 2313–2324 | 2319 View Article Online Environmental Science: Processes & Impacts Paper
Fig. 3 Dry–wet seasonal differences (“d” for dry, “w” for wet) of (a) discharge 105, (b) TSS, (c) temperature, and (d) E. coli in the Fanno, Johnson, and Balch Creek. Published on 14 July 2014. Downloaded by University of California - Davis 21/01/2016 01:40:41.
In addition, higher E. coli concentrations may be attributed Landscape impacts to the warmer temperature in the summer season, which According to the Mann–Whitney U test, all variables surveyed corresponds to higher growth and survival rates of E. coli (discharge, temperature, TSS, and E. coli)demonstrated bacteria.9,13 As E. coli bacteria are thermotolerant (tolerant of signicant differences across the Fanno, Johnson, and Balch relatively high temperatures), the lower river temperatures Creek watersheds at the 0.01 level (2-tailed), except for TSS during the cool season inhibit the growth and survival of E. coli difference between the Johnson and Balch Creek watershed. bacteria in the river. The reproduction of E. coli outside the The Johnson Creek has higher ow per drainage area than intestines of warm-blooded animals seems unlikely. However, the Fanno Creek despite having a lower degree of impervi- there is a growing body of evidence suggesting that there exists a ousness. This is probably because storm pipes reroute water specialized subset of E. coli strains that can reproduce in further downstream to the mouth of the Fanno Creek where a secondary environments in both tropical43,44 and temperate wastewater discharge plant is located. The Fanno Creek has climates.45–49 Therefore, the use of E. coli as an indicator of fecal the highest levels of TSS, E. coli, and temperature, followed by pollution should be reevaluated.50 the Johnson and Balch Creek in order (Fig. 3). The results are We were unable to identify a signicant seasonal pattern for expected since most pollutants could be attributed largely to TSS, indicating that either no seasonal difference exists for the anthropogenic sources. ffi study period, or there are insu cient data points to identify a Afewsignicant correlations were observed between ff signi cant di erence. It is possible that although the wet water quality and landscape variables (Fig. 4). Correlations season generates more runoff to carry sediment into streams, it between %IMP and E. coli in the wet season, as well as %DEV also has more dilution effect which might cancel out the addi- with TSS indicate that high impervious surface coverage and tional input. degree of urban development carry increased concentrations
2320 | Environ. Sci.: Processes Impacts,2014,16, 2313–2324 This journal is © The Royal Society of Chemistry 2014 View Article Online Paper Environmental Science: Processes & Impacts
of TSS and E. coli bacteria to surface waters. The results generally agree with a growing body of the scientic literature that predicts water quality degradation resulting from urbanization.5 However, most of the correlations are not signicant, suggesting that more study is needed to identify the underlying controls of transport and build-up of E. coli in water systems under various landscape regimes. We went further to analyze the response of the correlation between TSS and E. coli to landscape characteristics. Consistent with our hypothesis, a correlation between TSS Fig. 5 Probability plot (Q–Q plot) of % developed land. and E. coli signicantly decreased as %DEV went up, albeit to a small extent (Table 7). The best model was listed as follows: Cor ¼ season + %DEV. The whole model is signicant at the 0.05 level with a R2 of 39%, and both independent variables, Multiple regression analysis season and %DEV, are signicant at the 0.05 level (Table 7). We ran several multiple linear regression models to predict the Large variation was observed in regions with high percentage E. coli concentration based on the stepwise method and the of development, shown as the deviation at the upper right on knowledge gained from our correlation analyses. The Balch the normal Q–Q plot (Fig. 5). Most of those regions lie within Creek was excluded from the analysis since no signicant the Fanno Creek watershed, implying the more complex predictors were found within our targeted variables (antecedent stormwater infrastructure and more variable pollutant sour- precipitation, discharge, TSS, and water temperature). Table 8 ces of E. coli in the urban watershed.4 No signicant impact, summarizes the major information regarding our models. however, was identied in %IMP, suggesting complex Models for the Johnson Creek watershed (eqn (2) and (4)) have transport and build-up processes of E. coli and TSS, which much higher adjusted R2 than those for the Fanno Creek could hardly be represented as a linear function of degree of watershed, which are capable of explaining 60% and 50% of the imperviousness. Those processes include changes in the variability in the E. coli concentration for the dry and wet stream route caused by urban storm drains in the Fanno season, respectively. Again, the lower predictability of the Fanno Creek. %Forest also did not show any signicant impact on E. Creek models are likely the result of a more complex stormwater coli and TSS correlation. Apart from the trend distortion infrastructure and more variable pollutant sources of E. coli in caused by oversimplication, a possible explanation could be the urban watershed. that our sample size is too small to capture the trend since we Best models for the Fanno, and Johnson Creek watersheds only have eight data points. are listed as follows:
Fanno dry: ln(E. coli) ¼ 0.522 P1 (1)
Johnson dry: ln(E. coli) ¼ 0.735 TSS + 0.212 T (2) Published on 14 July 2014. Downloaded by University of California - Davis 21/01/2016 01:40:41. Fanno wet: ln(E. coli) ¼ 0.356 P3 + 0.309 P1 (3)
Johnson wet: ln(E. coli) ¼ 0.448 P1 + 0.528 TSS 0.305 P7 (4)
where P1, P3, P7, T, and TSS stand for the one-day, three-day, and seven-day antecedent precipitation, temperature, and total suspended solids aer standardization. Variables are ordered by coefficient signicance, and all of these selected variables are Fig. 4 Relationship between E. coli, TSS and landscape and the cor- signicant at the 0.05 level. The models for the Fanno Creek responding Spearman's rank correlation coefficients (two-tailed) for (a) watershed show positive relationships between the E. coli %DEV vs. TSS in the dry season, and (b) %IMP vs. E. coli in the wet concentration and the one-day, three-day antecedent season; both correlation results are significant at the 0.05 level.
Table 7 Landscape model summary (n ¼ 22)a
Estimate Std. error t value P-value Durbin–Watson